Why Rock Slope Stability Is Still Hidden From View, and Why It Matters

Engineers can now scan a cliff from orbit, fly drones along a pit wall and model a slope in three dimensions. Yet the thing that decides whether rock slope stability holds or fails is still hidden inside the rock. A single borehole or a clean exposed face can give you a picture of the wider slope that is far more reassuring than the ground deserves.

That gap matters more now than it did a decade ago. Rainfall patterns are shifting, monitoring tools are spreading fast, and artificial intelligence is being promoted as a fix. Dr Harry Holmes, an engineering geologist at the University of Leeds, argues the real core challenge is uncertainty itself.

For you as a mining or infrastructure investor, that uncertainty has a price. Pit walls, road cuttings and waste dumps that behave differently from the model can mean flatter designs, lost ore and production downgrades.

Here is where the uncertainty actually sits, what climate change alters, and which claims about monitoring and AI deserve your scepticism.

Why is hidden rock-mass variability still the central problem?

You might assume that more survey data means less risk. It is a reasonable belief, and it is only half right.

Nobody can see inside a slope directly. Engineers infer what is happening underground from what they can observe at the surface, then build a design from that inference. Holmes, speaking to GeoDrilling International in an article by Signe Hansen, puts it plainly:

The main challenge Managing uncertainty and the heterogeneity of the rock mass is, in Holmes’s view, the central difficulty in rock slope engineering.

Heterogeneity simply means the rock is not uniform. According to Holmes, the features that control failure can change sharply over just a few metres:

• Fracture orientation: the angle and direction of cracks in the rock
• Persistence: how far each crack extends
• Spacing: how close the cracks sit to one another
• Infill: the material packed inside cracks, such as clay
• Groundwater: where water sits and how much pressure it exerts
• Weathering: how far the rock has broken down chemically or physically

So the question is rather whether your data captured the specific discontinuities (breaks or weaknesses in the rock) that will govern failure. That is why old tools such as stereonets, which plot fracture orientations on a circular chart, and scanline surveys, which record every fracture crossing a measured line, remain in daily use alongside on-site inspection.

Water sharpens the problem. Guidance attributed to the British Geological Survey (BGS) notes that water adds weight, reduces friction and weakens material, meaning a slope can fail after rain even when its shape has not changed at all.

Scale magnifies everything. The CSIRO-linked Large Open Pit project highlights that big pits expose more rock over longer mine lives, so a small error in structure, groundwater or strength assumptions carries a much larger financial bill.

Why a single factor of safety can mislead

A factor of safety is the ratio of the forces holding a slope up to the forces pulling it down; above 1.0 suggests stability. Deterministic methods produce one such number by reducing a variable rock mass to a handful of representative values, which can quietly hide the spread of possible outcomes.

Probabilistic and observational approaches exist to manage that spread rather than pretend it away. Steve Rogers of WSP has reportedly discussed combining discrete-fracture-network modelling, which represents rock as a web of individual fractures, with limit-equilibrium analysis.

The lesson for you is direct. A sound resource estimate does not guarantee a deliverable mine plan, because orebody uncertainty and geotechnical uncertainty are separate risks.

Frameworks for geotechnical risk management treat slope behaviour as its own hazard category with dedicated controls, rather than folding it into a general resource model.

How is a changing climate altering the design basis for slopes?

Start with the reassuring part. Holmes expects soil nails, rock bolts and netting to change little, with any effects of climate emerging gradually and solutions adapting in step.

The sharper question is whether the historical rainfall record remains a safe design input. Five mechanisms explain why that question is being asked worldwide:

1. Pore-water pressure: rain soaks into joints and benches, raising water pressure and reducing the rock’s resistance to sliding.
2. Freeze-thaw and thermal stressing: water expands as it freezes in cracks, and repeated heating and cooling may also loosen rock over time.
3. Drying followed by intense rain: shrinkage cracks open in dry spells, giving later downpours a fast route into the slope.
4. Permafrost thaw and glacier retreat: loss of frozen ground or ice support destabilises mountain and high-latitude slopes.
5. Drainage overload: heavier runoff erodes slope toes, overwhelms ditches and lets water pond behind engineered structures.

The UK offers a well-documented example, though the mechanisms apply globally.

Treat the long-dated figures as scenarios, not current conditions.

The debate over historical design bases

Nobody seriously disputes that water destabilises slopes. The argument is over whether a storm that historical records call a one-in-a-hundred-year event still carries that probability, given the observed shift in heavy rainfall.

The counter-view is equally reasonable. Local geology, drainage, vegetation and groundwater connections may dominate a slope’s response, and short rainfall records carry high natural variability.

The defensible middle ground is to stress-test historical data rather than throw it out, adding updated rainfall projections, transient groundwater scenarios and compound dry-wet or freeze-thaw cases. Holmes expects the practical response to be extra drainage, shorter maintenance intervals and more frequent clearing behind netting.

Heavier runoff overwhelming ditches and ponding behind structures is the same pattern seen in mine water management failures, where drainage capacity sized to older rainfall records proves inadequate once intense events arrive.

For you, climate risk to a slope looks mostly like a gradual creep in drainage, maintenance and monitoring costs rather than a sudden change in engineering methods.

What does the Folkestone project show about turning theory into a working design?

The Lower Leas Coastal Park sits below a weathered sandstone slope in Folkestone, Kent, with a playground at its foot. Coastal storms and erosion had been loosening the face for years, and heavy rain in early 2024 brought slope movement and cracks in the walkway above.

The plan looked tidy. Tony Gee & Partners designed a hybrid system: rock bolts with stainless-steel mesh on the upper section and passive drape netting, which hangs over the face to guide falling rock downwards, on the lower section. CAN Geotechnical installed it.

The specifications come from supplier Geobrugg‘s project record. The two zones add up to about 23 m of a roughly 35 m slope, and the available sources do not explain the remainder.

Then the site intervened. Dense vegetation had prevented accurate early mapping of the topography, so Tony Gee’s engineers made regular visits with the contractor to shift nail positions to match the real ground.

Designs must be installable Holmes warns that a neat bolt pattern on a drawing may be impossible to achieve on site, which renders the design pointless.

Stabilisation was reportedly completed before the playground refurbishment began in April 2026. Even this modest, well-documented scheme needed field adaptation, so if a project budget treats the design as fixed at drawing stage, you should read it as understating risk.

Can monitoring, probabilistic modelling and AI actually remove the uncertainty?

The tools deserve real credit first. Remote sensing has expanded sharply over the past decade, and at West Bay in Dorset repeated LiDAR (laser scanning) and drone surveys track cliff change and rockfall volumes over time.

Repeated LiDAR terrain monitoring is now routine at mines, producing millimetre-scale surface models that let geotechnical teams compare successive scans and quantify rockfall volumes and wall displacement.

InSAR, short for interferometric synthetic aperture radar, uses repeated satellite radar images to detect ground movement of millimetres. That data can feed back into models and help set trigger levels for action.

Where monitoring gave warning

At Kuzbass coal pits in Russia, ground-based radar reportedly recorded 23 abnormal acceleration episodes, 19 of them followed by destructive events, with 82.6% detection accuracy and an average lead time of 67.4 hours. These figures are unverified, but they show what high-frequency monitoring can offer when it connects to a response plan.

At Vedanta Zinc International‘s Gamsberg Mine in South Africa, conference material describes InSAR and 3D modelling working as a feedback loop, with deformation trends proving informative even below formal alarm thresholds. At SSR Mining‘s Çöpler mine in Türkiye, retrospective InSAR reportedly found about 60 mm/year of displacement before the February 2024 heap-leach collapse.

That last example cuts both ways. A signal found after the fact is only useful if someone reads it in time and has the authority to act.

Monitoring can still fail because:

• radar and InSAR depend on line-of-sight geometry and can be blinded by shadowing
• vegetation obscures LiDAR, drone and radar returns
• atmospheric noise and decorrelation degrade satellite data
• a feature can accelerate between observations
• instruments may not distinguish seasonal movement from progressive failure
• a warning achieves nothing without intervention authority

Where AI and probabilistic methods fit

Holmes sees probabilistic and fracture-network methods as credible but time-intensive, mainly practical for mega-projects. He expects AI to have its greatest impact over the next five to ten years, helping refine designs and run analyses faster, while engineers still catch unworkable output.

Reviews published in 2025 and 2026 (unverified) flag that models performing well on one mine’s data may not generalise, and that data scarcity, interpretability and sensor reliability slow adoption. When a company announcement promises an autonomous “stable or unstable” answer, you should look for uncertainty bounds, physical reasoning and human sign-off before taking it at face value.

Treating slope risk as a cash-flow question, not a safety footnote

Uncertainty in rock slopes is structural, not a data shortage. Climate shifts raise the bar on design assumptions gradually, and monitoring and AI narrow the risk without closing it.

The consequences are already visible. Orla Mining reportedly cut 2025 production guidance at Camino Rojo from 280,000-300,000 oz to 265,000-285,000 oz after a pit-wall incident, and Çöpler’s collapse suspended the operation.

When you assess a project, look for:

• independent geotechnical review
• transparent structural models
• updated rainfall and groundwater assumptions
• reconciled monitoring data
• defined trigger-action-response plans
• contingency for geometry changes

Over the next five to ten years, watch whether AI tools prove they generalise across sites and whether operators publish how warnings translate into action.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Climate projections and technology forecasts are speculative and subject to change.